Feature Selection and Cancer Classification via Sparse Logistic Regression with the Hybrid L1/2 +2 Regularization

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Cancer classification and gene selection are crucial for genomic data analysis.
• Logistic regression is a popular classification method but lacks inherent feature selection capabilities.

Purpose of the Study:

• To develop a novel hybrid regularization function for effective gene selection in logistic regression models.
• To enhance knowledge discovery in genomic data by identifying relevant genes for cancer classification.

Main Methods:

• Proposed a hybrid L1/2 + L2 regularization (HLR) function combining sparsity and grouping effects.
• Developed a univariate HLR thresholding approach for coefficient updates.
• Implemented a coordinate descent algorithm for the penalized logistic regression model.

Main Results:

• The HLR approach effectively performs gene selection within logistic regression.
• The method demonstrates competitive performance compared to existing state-of-the-art techniques.
• Simulations and empirical results validate the efficacy of the proposed HLR method.

Conclusions:

• The novel HLR regularization offers a robust solution for gene selection in cancer classification.
• This method improves feature selection in genomic data analysis, aiding knowledge discovery.
• The developed algorithm provides a competitive and efficient tool for researchers in the field.